Tree density

Comparing outbreak dynamics and outcomes on reforestation and natural forest sites provided a framework for evaluating many of our hypotheses but representing the two habitat types with only two sites each (due to logistical constraints) limited our ability to statistically test our results. Nevertheless, large differences in tree species dominance between the habitat types allowed us to qualitatively evaluate the resource concentration hypothesis and our predictions that in the dense reforestation stands koa moths would build up numbers earlier and to higher levels, defoliation would be more severe and refoliation would be slower, and deposits of caterpillar frass and its resulting nutrients would be greater.

To quantify structural differences between reforestation and natural forest stands, we established 55 survey plots, 85% of which were in natural forest habitat. The total area surveyed was 17,279 m² (1.73 ha) of which forest habitat comprised 14,765 m² (Mix2 6911 m², Mix1 7854 m²) and reforestation habitat comprised 2513 m² (Koa1 1571 m², Koa2 942 m²). More area was surveyed in natural forest habitat due to its lower koa density and greater habitat heterogeneity. In each habitat, plots were placed at 50-m intervals that were offset at random distances and directions perpendicular to our line of travel along unimproved roads and fence lines. We marked each plot location using Global Positioning System (GPS) and counted all koa and ‘ōhi‘a within a 10-m radius (314 m²) of the plot center. Densities of koa and ‘ōhi‘a were estimated from the count data pooled from all plots at each site.

Tree size

Koa trees usually (54%) consisted of one main stem at 1.4 m height, but dbh for trees with two (29% of trees) or more (17%) major stems was measured at 1.4 m height at each stem (a, b, c, …) and the total was calculated with the formula (√[a² + b² + c² + ···]; USFS, 2007). We placed ‘ōhi‘a into size classes 1–3 (as above), but we did not measure dbh except when we could not determine size class by visual inspection.

Moth abundance

At Hakalau, we sampled the dynamics of moth abundance with Townes-style ground and canopy malaise traps (BioQuip Products, Rancho Dominguez, California, and Sante Traps, Lexington, Kentucky, respectively) at each site except at Koa1, where only a ground trap was installed due to logistical constraints. We deployed ground and canopy traps at both lower- and higher-elevation trap locations at Mix2. Ground traps were deployed on 11 April 2013 and canopy traps deployed on 16 May 2013 and were operated continuously until 26 September 2013. Contents of traps were collected every 2 or 3 weeks, and moths were counted according to species using reference specimens.

At Laupāhoehoe study sites, we installed Townes-style ground malaise traps but not canopy traps.

Caterpillar abundance, biomass, and age structure

We counted caterpillars on 4–10 branches removed from up to 10 koa trees at each site at 1 to 3-week intervals from 18 April to 26 July 2013 (11 sampling events). Samples from the two natural forest trapping locations at Mix2 were composited for caterpillar assessment. Branches were randomly sampled before and during the early stages of the outbreak, but as foliage became patchy and sparse, we targeted branches containing sufficient foliage to support caterpillars. We used one extendable pole to clip a branch while a second pole with a nylon bag was held beneath to catch it. Each branch sample was approximately 50 cm in length. Caterpillars were separated from samples by shaking the branches inside the bag, after which the head width of each caterpillar was measured to determine its instar category (1–5). Foliage was separated from stems and dried to a constant mass at 50°C. One subset of caterpillars was maintained on young koa phyllodes to measure rates of parasitism (see below) and to confirm species identification. Another subset of caterpillars was dried to a constant mass at 50°C to determine the average mass for each of the five instars, which was then used to calculate the biomass of all caterpillars within the sample.

Patterns of defoliation and refoliation

We predicted that the high-density reforestation stands would be defoliated earlier and more heavily and that refoliation would be slower in the more heavily defoliated trees. At each marked koa, therefore, we visually estimated the amount of foliage that had been eaten by caterpillars from 0% to 100% in 5% increments (also see Nakajima, 2018). Any uneaten foliage was quantified to the nearest 5% according to the proportion of leaves or phyllodes. The amount of new foliage produced after defoliation was assessed from September through December, when most trees were recovering vigorously. New foliage was quantified to the nearest 5% according to the proportion of leaves or phyllodes.

We used a generalized linear modeling approach with Akaike information criterion with correction for small sample size (AIC_c) model selection to assess the effects of the severity of defoliation, tree size (dbh), habitat type (natural forest, restoration stand), and site on the degree of new foliage production 34 weeks after the peak of defoliation.

Litter traps

We deployed eight litter traps under koa canopies at each of two natural forest sites (Mix1 and Mix2) and one reforestation site (Koa2) to measure litterfall rates during and after defoliation. Traps were made from plastic lattice seedling flats (41 × 41 cm) with mesh screen glued to the bottom and sides. Sampling started in May 2013, and we collected litter monthly until August 2014. Litter types were not separated by species, although most litter was from koa. Large branches, twigs, and loose lichen material that had presumably fallen from the canopy were discarded because phyllodes were most likely to contribute to a nutrient pulse, given the N-fixing ability of koa, and other material would not contribute to nutrient cycling during the time frame of our study. Litter was dried to a constant mass at 70°C.

Nitrogen redistribution

We estimated N redistribution rates to evaluate variation in N pulse data derived from estimates of caterpillars (see above), frass production, and foliage consumption. For each sample date, we multiplied the number of caterpillars of each instar (individual/g foliage) by frass production of each instar (mean dry mg frass/caterpillar) to estimate frass production (dry mg frass/g foliage). We used foliage biomass (g/m) to predict frass deposition per unit area and sample date. Lastly, we linearly interpolated between caterpillar sample dates to estimate frass production over time.

Koa foliar biomass under non-outbreak conditions was calculated by allometric equations (Pearson & Vitousek, 2001) for dbh 1–8 cm (y = 0.0268 × dbh^2.168) and for dbh 8–30 cm (y = 0.034 × dbh^1.746). The equation for small koa was developed for dbh 1.5–8 cm. The equation for large koa was developed for dbh 8–30 cm, but we also used it for dbh >30 cm (10% of trees had dbh 30.1–60.0 cm; 3% of trees had dbh 60.1–300 cm). Improved methods for estimating the foliar biomass of koa (Ostertag et al., 2014) require measurements of tree height, which we did not record. Nevertheless, the relatively similar ratios of small to large trees at natural (47:53) and reforestation (58:42) stands indicate that estimates of foliar biomass are comparable.

Caterpillar frass production was estimated by rearing caterpillars in growth chambers maintained at 15°C, which approximated the daily average (12.2–13.8°C) and hourly maximum (14.9–16.4°C) temperatures expected at our Hakalau study sites during June (Giambelluca et al., 2014). Caterpillars representing instars 3, 4, and 5 were weighed at the beginning of the experiment after having been fasted for 24 h and then fed fresh, young koa phyllodes over a 48-h period. Frass produced by these caterpillars was collected at 24-h intervals and subsequently dried to a constant mass at 50°C. Caterpillars were reared to moths after the experiment to confirm their identities. Mean dry mass of each of the five instars was calculated for caterpillars collected during the outbreak (i.e., n = 30 randomly selected individuals per instar). A best-fit regression equation based on the amount of frass produced by instars 3–5 was used to estimate frass production for instars 1 and 2, where instar mass was used to predict frass mass.

We pooled samples of frass from caterpillars in the laboratory (see above) into a single aggregate that was dried (70°C) to a constant mass, ground, subsampled in triplicate, and analyzed for total %N by the University of Hawai‘i, Hilo Analytical Lab. To determine whether there was a distinct peak in frass production over time, we used the fit curve platform in program JMP (version 10.0.2; 2012; SAS Institute, Cary, North Carolina), using a Gaussian Peak function and site as the level of replication. Frass collected from randomly selected caterpillars reared in the laboratory on young koa phyllodes was composited into a single sample for determination of total N by the University of Hawai‘i, Hilo Analytical Laboratory to be used in N deposition estimates.

Resin bags

Variation in available soil N and P over time was measured with ion-exchange resin bags. Resin bags capture ions present in soil solution and are well correlated with other measures of inorganic nitrogen (Binkley & Matson, 1983) and effectively capture increased P flux to soil (Shaw & DeForest, 2013). Resin bags sorb nutrients throughout their time of burial in soil and thus can provide a more integrated measure of overall nutrient flux than static soil nutrient tests (Binkley & Matson, 1983). We refer to the inorganic nitrogen captured by resin bags over the time of burial as “resin available N.”

We made resin bags by sewing 8 g of wet Mixed Bed Anion Exchange IONAC Resin (J. T. Baker, Center Valley, Pennsylvania) in silkscreen pouches. These were buried at an angle at a depth between 5 and 10 cm, using a soil knife to lift soil, creating as little soil disturbance as possible. We used a shallow depth to quickly capture the N signal from the caterpillar frass as it moved down the soil profile. Resin bags were buried under the koa canopy and in adjacent open-canopy grassland plots. The understory of koa also tended to be occupied by exotic grasses, although understories of certain intact forest sites were also occupied by native shrubs. We placed resin bags within eight paired plots at the two natural forest sites and one reforestation site, for a total of 48 resin bags per time point.

Resin bags were buried between 15 and 47 days spanning May 2013 to May 2014. Caterpillars were already present during the first resin bag deployment, and, due to observations of heavy frass deposition, we collected the bags only 15 days later. We accounted for varying burial times by calculating resin available N as a rate per day. We continued collecting and replacing resin bags every month for 240 days, after which we skipped every third month. Between time points, we changed resin bag placement slightly to avoid disturbed soil. Resin bags were returned to the laboratory, rinsed with deionized water to remove roots and soil, and extracted by shaking with 100 mL 2 M KCl for 6 h. Samples were analyzed for ammonium (NH₄⁺) and nitrate (NO₃⁻) at the University of Hawai‘i, Hilo Analytical Lab on a Lachat Qui Chem 8500 (Hach, Loveland, Colorado) and for soluble reactive P on a Lachat QC8000 at the U.S. Geological Survey Forest and Rangeland Ecosystem Science Center (FRESC) laboratory in Corvallis, Oregon. We used repeated-measures analysis with site and habitat (koa vs. grass) as fixed effects. We log-transformed data to account for non-normality.

Understory species foliar nitrogen

To evaluate patterns of N uptake by understory vegetation, we first measured the percent cover of understory species in natural forest plots to characterize understory community composition. At one reforestation site, Koa2, we added five transects spaced at 50-m intervals. Few plots were established at this site due to the high density of koa and homogeneity of the canopy, but additional replication was needed to characterize the understory. We used plot centers as the midpoint of a 20-m transect that followed a north–south bearing and used a point-intercept method by holding a 2-m pole upright every 0.5 m. We noted all species touching the pole below 1 m, and we included litter or moss at the bottom point of the pole. Percent cover was calculated as the number of contact points divided by the total number of sample points (40). Because all plant species touching the pole at each point were recorded, percent cover values often summed to >100%, reflecting the multi-layered nature of the understory. We used linear discriminant analysis to cluster sites, given the percent cover of understory species.

We tracked the foliar N of species in open areas and under the koa canopy to determine whether understory plants displayed increased N uptake due to frass deposition. We sampled four of the eight plots that contained resin bags in each site by taking new, fully expanded leaves from exotic pasture grasses and from the native woody species ‘ōhi‘a, ‘ōhelo (Vaccinium calycinum), and pūkiawe (Leptocophylla tameiameiae). We pooled two to three leaves, including petioles, from three separate individuals (or bunches in the case of grasses) per replicate plot. Grasses were a mixed-species assemblage dominated by Cenchrus clandestinus, but also including Anthoxanthum odoratum, Ehrharta stipoides, Holcus lanatus, Paspalum dilatatum, Agrostis stolonifera, and Axonopus fissifolius. We took care not to include any inflorescences in the grass foliage samples. Foliage was sampled in May and September 2013 and January 2014. We added one extra sample date for grasses (August 2013) as grasses tend to grow and assimilate resources more quickly than woody species. Foliage was dried to a constant mass at 70°C, ground with a Wiley Mill (Thomas Scientific, Swedesboro, New Jersey), and analyzed for total carbon (C) and N by the University of Hawai‘i, Hilo Analytical Lab (Costech 4010 Elemental Analyzer, Valencia, California). Foliar samples were not analyzed for P.

Use of tree species

Anticipating that birds, regardless of their foraging guild, would be attracted to a pulse of high-quality insect prey, we monitored the relative frequency of birds in koa and ‘ōhi‘a trees during the outbreak. In the reforestation plots, we expected little change in behavior because birds were expected to continue to forage predominantly in koa, due to its dominance there. In the natural forest plots, however, we expected to see birds spending more time in koa and less time in ‘ōhi‘a, although we anticipated that ‘ōhi‘a flowers would continue to attract nectarivores, possibly diminishing their attraction to koa compared to insectivores.

To evaluate the effects of forest structure on foraging behavior, we conducted 1017 focal surveys of koa and ‘ōhi‘a trees in each of the four study sites during 12 surveys at approximately weekly intervals between 22 May and 15 August 2013. We estimated the date of peak defoliation as 20 June for the two reforestation sites and the Mix1 natural forest site and 27 June for the Mix2 natural forest site. We estimated heights of trees to the nearest meter. Mean heights of koa and ‘ōhi‘a were similar within the natural forest sites (14.7–14.9-m koa; 14.8–16.0-m ‘ōhi‘a) and within the reforestation sites (10.7–11.4-m koa; 11.8–11.9-m ‘ōhi‘a), but both species were 3–4 m shorter in reforestation sites compared to natural sites. We estimated flower production of ‘ōhi‘a as: none (0); light (≤5% of tree canopy with flowers); medium (≤20%); heavy (>20%). We classified the extent of defoliation of koa as: none, light (1%–25% of foliage missing), or heavy (>25% of foliage missing). We counted the number of birds that visited koa and ‘ōhi‘a trees during a 2-min period. The presence of a bird in a tree was recorded regardless of its activity there; for example, a bird was counted if it perched even briefly in the tree. We attempted, but often failed, to identify the species of birds occupying the tree. As an index of bird activity, we calculated the number of birds per min; this calculation included instances when we observed no birds in a tree. Focal trees (usually >5 m in height) were selected without regard to phenology as we walked along or near roads or fence lines.

We used logistic regression to identify factors explaining observed differences in the proportion of birds observed in koa and ‘ōhi‘a before and after peak defoliation in both the natural forest and reforestation habitats. Predictor variables were tree species (koa, ‘ōhi‘a), habitat type (natural forest, restoration), date, severity of koa crown defoliation (as above), ‘ōhi‘a flowering intensity (as above), and observer. The analysis categorized the proportion of birds in either koa (with or without defoliation) or ‘ōhi‘a (with or without flowers). To simplify the interpretation of coefficients, some variables were combined. Flowdefol represented a combination (factor plus interaction) of the three categories of ‘ōhi‘a flowering plus the three categories of koa defoliation. Sppflow represented a combination of koa (regardless of defoliation status) plus the three categories of ‘ōhi‘a flowering. Sppdefol represented a combination of ‘ōhi‘a (regardless of flowering status) plus the three categories of koa defoliation. We analyzed data in this way for all species combined and for the four most frequently observed species: ‘apapane (APAP, Himatione sanguinea), Hawai‘i ‘amakihi (HAAM, Chlorodrepanis virens), ‘i‘iwi (IIWI, Drepanis coccinea), and warbling [Japanese] white-eye (WAWE, Zosterops japonicus).

Changes in diet

To investigate whether birds shifted their diets to include more koa moth larvae in response to the outbreak, we identified arthropod prey fragments within fecal samples. Birds were captured and banded in 2013 at Koa1 (reforestation) and Mix1 (natural forest) as part of a multi-year demographic study of forest birds begun in 2012. Work began 26 February as part of the pre-planned banding season, which was scheduled to run for 3 months during the core breeding season. However, in response to the koa moth outbreak, we extended the banding season until 27 June. From February through April, each site was visited three times during every 2-week period. During May and June, we visited the two Pedro sites once each week. Mist nets (2.6 m × 12 m) at established locations (typically 10–12 nets/day at each site) were used to capture birds across the study period. Captured birds were removed from the nets, placed in light-weight cotton bags to keep them calm and safe, and taken back to a central banding location and held until ready for processing, weighing, and releasing. At the time of release, the bird bags were carefully searched for fecal samples deposited by the birds (bags were used only once per bird, then were cleaned). If a fecal sample was found, the sample was carefully scraped into a plastic microcentrifuge tube and immersed in 95% ethanol for storage.

Diet analysis focused on fecal samples collected opportunistically from Hawai‘i ‘amakihi and warbling white-eye, both of which feed on arthropods and ‘ōhi‘a nectar. Although these two species provided the most fecal samples for comparing pre-outbreak and outbreak diets, we also combined samples from forest and reforestation sites to increase our power to detect changes. Pre-outbreak samples were collected from 26 February to 25 April, before we observed the rapid increase in caterpillar numbers, and outbreak samples were collected from 22 May to 27 June, when caterpillars were abundant. Overall, 34 Hawai‘i ‘amakihi and 63 warbling white-eye samples were processed, resulting in 19 pre-outbreak and 15 outbreak samples for Hawai‘i ‘amakihi and 24 pre-outbreak and 35 outbreak samples for warbling white-eye.

In the laboratory, fecal samples were teased apart, and individual arthropod fragments were photographed and compared to a reference collection of known arthropods collected primarily at Hakalau. Estimates of numbers of individual arthropods within a sample were conservative, so for example, if three spider fangs similar in size and structure were found in a sample, it was assumed that there were two individuals in the sample, rather than three, because each spider has two fangs. The shape of caterpillar mandibles can indicate taxonomic affinity, although many mandibles were not identified to species, genus, or family level. However, the unique shapes of koa moth caterpillar mandibles are known and thus were quantified separately from all other taxa.

Changes in mass

We used mass as a measure of the change in fitness of birds in response to the caterpillar outbreak. Mass can reflect either the accumulation of fat (e.g., resulting in mass gain due to birds consuming more energy than they are burning) or the loss of muscle (and mass) that can be associated with insufficient food intake. We set 26 February to 21 May as the pre-outbreak period (25 banding-days) and 22 May to 30 June as the outbreak period (10 banding-days). A total of 656 birds were captured and processed at the two sites during this 4-month period.

We used an analysis of variance model with an adjustment for multiple comparisons (Holm, 1979) to test for significant differences (α = 0.05) in species mass pre-outbreak versus outbreak. Data were tested for normality, and only species with five or more captures in each of the two periods were included. Program JMP (version 10.0.2; 2012; SAS Institute, Cary, North Carolina) was used for statistical analysis.

Tree density

Our survey of 55 plots (1.73 ha total) confirmed that koa dominated in all size classes at reforestation sites (Table 2). Compared to koa in natural forest sites, reforestation sites supported about 2.5 times greater basal area (m²/ha), 8.6 times greater density of small trees (dbh 1–8 cm), and 4.8 times greater density of large trees (dbh >8 cm). Koa densities were highest at Koa2 in all size classes. ‘Ōhi‘a dominated the natural forest sites with a density of large trees (dbh >8 cm) twice that of large koa and 43.7 times that of ‘ōhi‘a in reforestation sites. Although large ‘ōhi‘a were nearly absent from the reforestation stands, they were scattered immediately outside of the planted areas and were frequently visited by birds from the koa stands. Additionally, the density of small ‘ōhi‘a (dbh 1–8 cm) was slightly greater in the reforestation sites. The Koa2 reforestation site was exceptional in terms of its high densities of both small koa and ‘ōhi‘a.

Understory community

Understory communities differed in species composition between sites. Koa1 and Koa2, the two reforestation sites, were dominated by exotic species more than were Mix1 or Mix2, largely due to the prevalence of exotic grasses (Appendix S1: Tables S1 and S2). In contrast, vegetative cover of native shrubs, ferns, and trees was higher at Mix1 and Mix2. Discriminant analysis indicated that Koa1 and Koa2 were more like one another than they were to Mix1 or Mix2 (Appendix S1: Figure S1). Although the latter two sites had relatively high percentages of native species cover in the understory, they separated in multivariate space due to different native species being present or dominant.

Patterns of defoliation

Defoliation severity was variable at the four sites, and the pattern did not meet our expectation that reforestation sites would be more heavily defoliated than natural forest sites. The amount of foliage collected during branch clipping decreased over time until maximum defoliation on 20 June at the two reforestation sites and the Mix1 natural site and 27 June at the Mix2 natural forest site. Compared to pre-outbreak levels, foliage biomass declined by 69% overall, with natural sites experiencing relatively high levels of defoliation (72% at Mix2, 85% at Mix1). Defoliation was also high at one of the reforestation sites (82% at Koa2) but was relatively low (42%) at Koa1 by 13 June.

At landscape scale, defoliation was heavy and widespread, averaging 55%–93% of crown volume across all size classes of koa in both habitats (Table 4). Severe defoliation (≥90% of crown) was observed among 57% of natural forest seedlings (dbh <1 cm) and 77% of reforestation seedlings as well as 83% of small trees (dbh 1–8 cm) in each habitat. More large trees (dbh >8 cm) were severely defoliated in natural forest sites than in reforestation sites (80% vs. 49%; χ² = 7.01, df = 1, p = 0.008). Among large trees, the extent of extreme defoliation was heaviest at the Mix1 natural forest site (94%), similar at both the Mix2 natural forest site (62%) and Koa2 reforestation site (61%), and least at the Koa1 reforestation site (39%).

Changes in foliar biomass

Foliar biomass, initially consisting almost entirely of phyllodes, was higher per unit of area in reforestation sites before defoliation (Table 5). Nevertheless, due to heavy defoliation at both natural forest sites but only one reforestation site, foliar biomass was reduced by about 78%–93% in natural forest habitat (two-tailed Mann-Whitney U compared to about 64%–79% in reforestation habitat test corrected for ties across groups; t = 4.07; p < 0.001). Phyllodes were nearly always consumed first, but we occasionally observed caterpillars eating true leaves on koa seedlings even when phyllodes were available. We also occasionally observed swarms of caterpillars on other plant species growing underneath koa, including pilo (Coprosma rhynchocarpa) and ‘ākala (Rubus hawaiensis). Although we observed herbivory on the leaves of these species, it was generally not extensive.

Response of koa to defoliation

Following defoliation, production rates of new foliage were variable and were not always consistent with our expectations. Few true leaves and phyllodes were evident 12 weeks after defoliation, but at 24 weeks, 35% of koa crowns (all size classes combined) consisted of new foliage. Small seedlings (height <1.4 m; no dbh) produced new foliage rapidly in both habitat types, but rates were slower for larger seedlings (dbh <1 cm) and small trees (dbh 1–8 cm; Appendix S1: Table S3). Large trees (dbh >8 cm) initially produced foliage more quickly in natural forest habitats, but foliage production in reforestation habitats had caught up by week 34. In both habitats, large trees produced about 1.5 times more new foliage (as a percentage of crown volume) than had small trees by week 34.

The amount of new foliage may also have been influenced by the severity of crown defoliation, although sample sizes for lightly defoliated koa were small. Seedlings and small trees (dbh ≤8 cm) that were relatively lightly defoliated (≤50%) tended to produce more new foliage than did seedlings and small trees that were more heavily defoliated, even after 34 weeks (Appendix S1: Table S3). Contrary to expectations, heavily defoliated large trees (dbh >8 cm) initially produced more new foliage than did lightly defoliated trees, but we recorded more foliage on lightly defoliated large trees than on heavily defoliated large trees after 34 weeks. Subsequently, we could no longer discriminate between new and old foliage, but the canopies even of heavily defoliated trees in both habitats had regrown most of their foliage after 1 year.

The best-fit model of foliage regrowth at 34 weeks included defoliation severity, tree size, and site as contributing factors. Taken individually, the severity of defoliation had a non-significant negative effect on regrowth (odds ratio of 90% reduction in regrowth for 100% defoliation), and dbh had a non-significant positive effect (10% increase in regrowth/10 cm of dbh). Tree mortality by week 34 was nearly entirely explained by tree size and severity of defoliation: 13 of 14 dead trees were small (dbh 1–8 cm), all were severely defoliated (99%–100%), and equal numbers died in each habitat type.

Of 510 seedlings, saplings, and small and large koa in our sample, 93 (18%) produced little or no (0%–5%) new foliage 24–25 weeks after the partial or complete loss of their canopies. Five trees and two seedlings that were moderately defoliated (10%–50%) recovered <5% of their canopies. Of trees that were heavily defoliated (>50%; n = 452), 40 (9%) produced 0%–5% new foliage 24–25 weeks after defoliation (Appendix S1: Table S3).

True leaves composed 73% of the new foliage across all koa size classes and habitats 24–25 weeks after defoliation. The new foliage of small trees (dbh ≤8 cm) was predominately true leaves in both natural forest (81%) and reforestation (89%) habitats, but true leaves composed only 55% of the new foliage of large trees in natural forest habitat and 32% in reforestation habitat.

Frass production

The amount of frass produced by caterpillars reared over 24 h at 15°C in the laboratory showed a strong relationship with caterpillar mass for instars 3, 4, and 5 (y = 2.2747x − 0.1595, R² = 0.9985; Table 6). Based on this equation, we estimated that individual first- and second-instar caterpillars produced <0.01 and 0.4698 mg of frass over 24 h, respectively. Therefore, the maximum daily frass production (kg/ha) of all caterpillars at the reforestation sites ranged from 92.63 (Koa1) to 280.28 (Koa2), and at the natural forest sites, it ranged from 100.80 (Mix1) to 206.63 (Mix2).

Litterfall

Initially, canopy litter fell at the highest rates at the Koa2 reforestation site, although litterfall dropped precipitously at this site after caterpillars reached maximum biomass in June (Figure 6). Litterfall rates at the other sites started to decrease ~1 month later. After defoliation, litterfall rates stabilized at generally lower levels at all sites. A peak in litterfall in late 2014 was due to a large wind event, which resulted in as much litter as during the outbreak.

Nitrogen concentrations were lower in koa phyllodes than in caterpillar frass (%N content: koa phyllodes [mean ± SE] = 2.46 ± 0.10, frass = 3.30 ± 0.02). Phyllodes also contained slightly lower carbon concentrations (%C content: koa phyllodes = 49.16 ± 0.36, frass = 51.30 ± 0.07), leading to lower overall C:N ratios for frass (C:N: koa phyllodes = 19.98 ± 0.77, frass = 15.55 ± 0.14).

Nitrogen redistribution

Cumulative frass and N redistribution were highest at the Koa2 reforestation site and the Mix2 natural forest site, where about 5600 kg/ha of frass (equivalent to about 180 kg N/ha) fell to the ground during about 2 months from April through July 2013 (Figures 1d and 7). High frass deposition in these two sites was due to high caterpillar density and high foliar biomass. At the two reforestation sites, where foliar biomass was similar (~500 g/m), caterpillar abundance was higher, and more frass fell at Koa2 than at Koa1. By comparison to N falling in frass, there was 160 ± 37 kg N/ha in the pre-defoliation koa canopy foliage at Koa2 and 86 ± 30 kg N/ha at Mix2.

Soil nitrogen (N) and phosphorus (P)

Resin available N in soils under koa canopies in both reforestation sites and the Mix1 natural forest site pulsed distinctly, while in open grass plots it did not, although this interaction was stronger in Koa2 than in Mix1 (Table 7, Figure 8). We did not detect a distinct N pulse at the Mix2 natural forest site. Nitrogen at the Mix1 natural forest site and the Koa2 reforestation site peaked once in summer (from 3 July to 19 August 2013) and again in winter (from about 23 October 2013 to February 2014), but with some variation in timing between sites (Figure 8). The midpoint of the first peak occurred ~43 days after maximum caterpillar biomass at Mix1 and Koa2. Soil resin N was higher on all dates in plots under koa compared to grass plots not under koa, and this difference was most pronounced when resin N was at its peak. In late March 2014, when the N pulse seemed to have dissipated, there were 12, 5, and 3 times as much resin N in koa plots as in grass plots in Koa2, Mix1, and Mix2, respectively. By contrast, at peak N for each site, there were 33, 19, and 4 times as much resin available N in koa plots as in grass plots in Koa2 (December 2013), Mix1 (October 2013), and Mix2 (May 2013). Available N under koa was first detected as both NH₄⁺ and NO₃⁻, although this quickly converted to an all nitrate signal 1 month after the initial pulse (data not shown).

For resin available soil P, only the Koa2 reforestation site had greater levels under koa than grass, and no differences were found at other sites. Similarly, soil P peaked distinctly under koa only at Koa2, where we observed a double peak that was similar to the double peak of N but offset by 30–60 days (Figure 8). Ratios of resin available N:P fluctuated at two sites, pulsing only at Koa2 toward the end of the study period (March 2014) and after the pulses in N and P alone.

Understory foliar response

Understory species responded variably to frass deposition (Figure 9). Grasses under koa responded strongly across all sites, with a 15% increase in foliar N from May through August 2013, ~1 month after soil N started to pulse. At the same time, the foliar N of grasses decreased by 21% in plots away from koa, leading to a significant time × habitat interaction in a repeated-measures analysis of variance (time effect, F = 10.73, p < 0.01; habitat effect, F = 4.17, p < 0.01; time × habitat interaction, F = 1.75, p < 0.01). Native woody species, ‘ōhi‘a, ‘ōhelo, and pūkiawe, also changed slightly in foliar N at this time, although time, habitat, or interaction effects in repeated-measures models were not significant.

General activity and response to habitat variables

Unidentified species composed 63% of the total number of birds observed (Table 8). Among the species we identified most frequently, ‘i‘iwi and ‘apapane were primarily in natural forest habitat, whereas Hawai‘i ‘amakihi and warbling white-eye were mainly in reforestation habitat. We often could not identify the behavior of birds in the focal trees, but we sometimes observed active foraging on caterpillars, which occasionally escaped predation by suddenly dropping from phyllodes on silk threads.

Logistic regression indicated that overall bird activity in tree species was influenced primarily by the degree of ‘ōhi‘a flowering, severity of koa defoliation, and observer (relatively few observers contributed most of the observations). These factors and habitat type also affected activity levels of the four most frequently observed bird species (Appendix S1: Table S4).

‘Ōhi‘a phenology was similar in natural forest and restoration habitats. About one-third of ‘ōhi‘a trees were flowerless before (33%, n = 100) and after (36%, n = 402) defoliation, but trees bore medium to heavy flower crops more frequently before (14%) compared to after (3%) defoliation (χ² = 21.47; df = 1, p < 0.001).

Changes in use of tree species

We expected birds in both habitats to be present more frequently in koa than ‘ōhi‘a during the buildup of caterpillars but then visit ‘ōhi‘a more frequently after peak defoliation. Instead, birds visited ‘ōhi‘a at least as often as koa before and after defoliation in both habitats. In natural forest habitat, birds were observed at similar frequencies in ‘ōhi‘a and koa before defoliation (‘ōhi‘a 57%, koa 44%; χ² = 2.77, df = 1, p = 0.096) and more frequently in ‘ōhi‘a after defoliation (‘ōhi‘a 39%, koa 14%; χ² = 33.79, df = 1, p < 0.001). In reforestation habitat, birds were present in ‘ōhi‘a at least twice as often before defoliation (‘ōhi‘a 66%, koa 33%; χ² = 8.64, df = 1, p = 0.003) and after defoliation (‘ōhi‘a 35%, koa 15%; χ² = 18.42, df = 1, p < 0.001). These results also indicate that the magnitude of change in the use of tree species varied according to habitat type. When caterpillars became scarce after defoliation, bird presence in ‘ōhi‘a declined from 57% to only 39% in natural forest but it fell from 66% to 35% in reforestation habitat. The magnitude of change in koa use indicated a reversed trend as bird frequency declined from 44% to 14% in natural forest and from 33% to only 15% in reforestation habitat after defoliation. Additionally, the significant drop in the use of both koa and ‘ōhi‘a after koa defoliation (Appendix S1: Table S5) may indicate post-defoliation dispersal to other areas not yet or less affected by the outbreak or a coincidental seasonal decrease in bird activity due to a change in other resources, such as ‘ōhi‘a flowers, as noted above.

In weak alignment with the general effect of tree species on bird presence or absence, bird activity in koa (both habitats combined) averaged 0.33 birds/min ± 0.05 SE before defoliation and 0.15 birds/min ± 0.03 after defoliation (z_1-tail = 1.26, p = 0.10). Bird activity was also twice as great in koa that had been defoliated lightly or not at all (≤25%; 0.20 birds/min ± 0.06 SE) compared to more severely defoliated koa (>25%; 0.10 birds/min ± 03; z_1-tail = 2.40, p = 0.008). Bird activity also declined in ‘ōhi‘a, averaging 0.74 birds/min ± 0.10 before and 0.39 birds/min ± 0.03 after koa defoliation (z_2-tail = 3.24, p < 0.001).

Changes in diet

As predicted from the pulse of high-quality prey, caterpillars increased in bird diets during the outbreak, but limited sampling prevented us from comparing diets of birds in different habitats or evaluating the diets of more than just two species. In the case of the native, omnivorous Hawai‘i ‘amakihi, we identified 689 individual prey from 19 fecal samples collected before the outbreak (from 26 February to 25 April) and 128 prey items from 15 samples collected during the outbreak (22 May–27 June). Hemiptera and Lepidoptera were the only taxa comprising >15% of prey items in both pre-outbreak and outbreak periods. Small-bodied Hemiptera, especially psyllids (Psyllidae) comprised 93% of prey items before the outbreak and 44% during the outbreak. Based on leg segments, juvenile wing buds, and adult wings, the primary psyllid prey species was the invasive Acacia psyllid (Acizzia uncatoides), itself an irruptive species on koa. Caterpillars, much larger prey than psyllids, comprised 4% of prey items before and 43% during the outbreak. Caterpillar fragments were found in 79% of samples before the outbreak and in all samples during the outbreak (χ² = 3.58, df = 1, p = 0.059). The number of individual caterpillars, as represented by paired sets of mandibles and unpaired mandibles, averaged 1.5/sample ± 0.28 SE before the outbreak and 3.7/sample ± 0.75 SE during the outbreak (t_1-tail = −2.55, df = 18, p = 0.008).

For the warbling white-eye, 796 individual prey from 28 samples were identified prior to the outbreak and 292 prey were identified from 35 samples during the outbreak. Psyllids and other Hemiptera comprised 87% of prey items before but 26% during the outbreak. Caterpillars comprised 5% of prey items before and 51% during the outbreak. Other taxa comprised <15% of the diet in both periods. Caterpillar parts were observed in 79% of samples before and in all samples during the outbreak (χ² = 8.29, df = 1, p = 0.004). The number of caterpillars in warbling white-eye samples averaged 1.6/sample ± 0.38 SE before and 4.3/sample ± 0.84 SE during the outbreak (t_1-tail = −2.91, df = 47, p = 0.003).

In the Hawai‘i ‘amakihi, koa moth caterpillars were identified in 19% of 19 samples before and 93% of 15 samples during the outbreak (χ² = 5.15, df = 1, p = 0.023). The abundance of koa moths was 0.32 caterpillars/sample ± 0.13 SE before and 3.0 caterpillars/sample ± 0.61 SE during the outbreak (t_1-tail = −4.30, df = 15, p = 0.0003). Koa moths were identified in 38% of 24 warbling white-eye samples before and 71% of 35 samples during the outbreak (χ² = 3.78, df = 1, p = 0.052). The abundance of koa moth caterpillars in warbling white-eye samples was 0.88 caterpillars/sample ± 0.36 SE before and 3.5 caterpillars/sample ± 0.88 SE during the outbreak (t_1-tail = −2.81, df = 15, p = 0.0036).

Changes in mass

Sampling limitations did not allow us to evaluate changes in mass by habitats, so weights were pooled across sites. A total of 573 individuals from six common forest bird species and the endangered Hawai‘i creeper (‘alawī; Loxops mana) were captured and measured at the Koa1 reforestation site and the Mix1 natural forest site in 2013 (Table 9). The mass of only one species (‘i‘iwi) was significantly different between the two sites, and the difference was consistent for pre-outbreak (26 February–21 May) and outbreak (22 May–30 June) periods, so sites were combined for analysis.

The more frequent consumption of caterpillars was expected to result in increased mass of birds generally. The direction and magnitude of change in pre-outbreak and outbreak mass varied among species and foraging guilds, but most species tended to gain mass (Table 9). Mass gain was unequivocal for the Hawai‘i ‘elepaio (Chasiempis sandwichensis) and Hawaii creeper. Arthropods are important in the diet of all species (Banko et al., 2015), but the Hawai‘i ‘elepaio and Hawai‘i creeper are primarily insectivorous whereas the Hawai‘i ‘amakihi and warbling white-eye are omnivorous (arthropods and nectar), the red-billed leiothrix (Leiothrix lutea) is frugivorous, and the ‘i‘iwi and ‘apapane are nectarivorous. Mass remained relatively unchanged during the outbreak only for the nectarivorous ‘i‘iwi and frugivorous red-billed leiothrix.

Response of birds to outbreak

Perkins (1903) noted that large numbers of birds were attracted to koa moth outbreaks, but we did not observe unusually high bird activity perhaps because caterpillars were distributed widely across the landscape. Bird activity was affected primarily by the abundance of ‘ōhi‘a flowers, severity of koa defoliation, and habitat type. Time spent in ‘ōhi‘a or koa changed in response to caterpillar abundance, but the direction of change differed according to habitat type. Additionally, birds increased their consumption of koa moth caterpillars as they became more abundant, and most species gained weight.

Birds occupied ‘ōhi‘a, which hosts many arthropods (Gruner, 2004) and is a major source of nectar (Banko & Banko, 2009), equally or more frequently than koa in both habitats, but the proportional use of tree species changed after defoliation, when both caterpillars and flowers became scarce. Relative to koa, bird activity increased in ‘ōhi‘a in natural forest habitat but declined in reforestation habitat after defoliation. As we expected, therefore, birds in natural forest habitat were attracted to koa when caterpillars were abundant, despite the relatively low density of koa. Contrary to our expectations, birds in reforestation habitat were more active in ‘ōhi‘a during the outbreak, despite the low density of ‘ōhi‘a. Although we have no supporting data, this suggests that birds might have readily obtain caterpillars in the high-density koa stands, allowing them more time to forage in ‘ōhi‘a for nectar and perhaps arthropods less available in koa.

Although caterpillars are the most frequent arthropod prey of birds generally at Hakalau (Banko et al., 2015), the availability of ‘ōhi‘a nectar strongly influences the foraging behavior and movements of nectarivorous species (Baldwin, 1953; Carothers, 1986; Carpenter & MacMillen, 1976; Pimm & Pimm, 1982), some of which likely dispersed as flowers became increasingly scarce during the outbreak. Many ‘i‘iwi disperse from Hakalau from May through August, when ‘ōhi‘a flowering typically decreases (Guillaumet et al., 2017; Kuntz, 2008), and in later months some ‘i‘iwi and ‘apapane, the most abundant species in natural forest habitat, move to subalpine habitat during the flowering peak of māmane (Sophora chrysophylla; Fabaceae; Hess et al., 2001). Hawai‘i ‘amakihi and warbling white-eye, the most abundant species in reforestation habitat, may have responded to changes in flower abundance with relatively localized movements. Birds spent less time in heavily defoliated trees compared to lightly defoliated trees, indicating that they were responding to reduced caterpillar availability. Birds also may have avoided defoliated trees to reduce their exposure to raptors, such as ‘io (Buteo solitarius). Although birds were affected by defoliation for a relatively short time at Hakalau, koa stands in the lower koa belt (e.g., Laupāhoehoe and Kīpuka) were defoliated repeatedly, with potentially greater impacts on regional bird populations. Whisson et al. (2018) found that Australian bird species richness was reduced and that birds used fewer microhabitats in defoliated trees, indicating that reduced resource availability can affect community structure, at least temporarily. Although we did not observe changes to the Hakalau bird community, native Plagithmysus (Cerambycidae) beetles may increasingly infest the many koa branches killed during defoliation (Goldsmith et al., 2007), and this could increase foraging opportunities for the endangered ‘akiapōlā‘au (Hemignathus wilsoni; Ralph & Fancy, 1996) and other insectivores.

The response of bird populations to insect outbreaks depends largely on the palatability of the irruptive insect (Paxton et al., 2011), and birds readily exploit some outbreaking insects (McMartin et al., 2002) or not (Hogstad, 2005). We expected birds to feed heavily on S. paludicola during the outbreak because Scotorythra spp. are important in the diets of Hawaiian birds generally (Banko & Banko, 2009; Perkins, 1903, 1913). Opportunistic species are expected to be especially prone to switching to a superabundance of palatable prey (Yang et al., 2008), and we found that two insectivorous/nectarivorous generalists, the Hawai‘i ‘amakihi and warbling white-eye, responded dramatically to the abundance of caterpillars during the outbreak. Both species shifted from a diet dominated by an invasive, irrupting psyllid before the koa moth outbreak to one dominated by koa moth caterpillars during the outbreak. Although the change may have been due to a seasonal decrease in psyllid abundance coinciding with the outbreak, the shift from other caterpillar taxa to S. paludicola, particularly by the Hawai‘i ‘amakihi, indicates opportunistic utilization of this temporarily abundant and valuable prey species.

In parallel with these dietary shifts, omnivorous and insectivorous bird species gained mass during the outbreak. Changes in mass were significantly positive for at least two (Hawai‘i ‘elepaio, endangered Hawai‘i creeper) of the four smallest and most insectivorous species (also including warbling white-eye, Hawai‘i ‘amakihi). Mass was unchanged for the frugivorous red-billed leiothrix and nectarivorous ‘i‘iwi, and mass apparently declined for the nectarivorous ‘apapane. Because sample sizes were small for ‘apapane, Hawai‘i ‘elepaio, Hawai‘i creeper, and ‘i‘iwi during the post-outbreak period, the influence of foraging guild or other factors potentially accounting for patterns of mass change is not clear.

Apart from increased mass, we were unable to evaluate other measures of fitness associated with the outbreak at Hakalau. The outbreak occurred during the later stages of the nesting season for most species (Ralph & Fancy, 1994; Woodworth & Pratt, 2009), when the abundance of highly nutritious food may have increased fledgling survival or possibly enabled additional nesting attempts. Nevertheless, following defoliation and the crash of caterpillars on koa, parent birds and fledglings would have had to rely more heavily on ‘ōhi‘a resources, especially in reforestation stands.

Predation by birds may reduce populations of foliar-feeding caterpillars under non-outbreak conditions (Barbaro & Battisti, 2011; Glen, 2004; Holmes, 1990), but it is ineffectual at preventing high densities (millions/ha) of caterpillars (Crawford & Jennings, 1989), levels that we observed during the peak of the outbreak. We identified caterpillars of Scotorythra spp. in fecal samples collected from 1994 to 1997 from Hakalau bird species (Banko et al., 2015), but if birds have a role in suppressing Scotorythra populations, it is likely restricted to the upper koa belt, where bird diversity and abundance, especially of insectivores, are much higher than at lower elevation (Camp et al., 2010; Scott et al., 1986). The single defoliation event at Hakalau raises the question of whether bird predation may have helped suppress subsequent irruptions, even if it failed to suppress the first. Caterpillars are more likely to be vulnerable to attack by predators, parasites, and pathogens at higher elevations, where larval development is slower, but slower larval development itself may have contributed to there being only one defoliation event.

Response of bats to outbreak

Where bats are diverse and abundant, they have been shown experimentally to limit arthropod populations and herbivory in natural and agroforestry habitats (Kalka et al., 2008; Kalka & Kalko, 2006; Williams-Guillén et al., 2008). Given the relatively low abundance of the Hawaiian hoary bat, it seems doubtful that they suppress insect populations in native forests or play an important role in the dynamics of insect outbreaks. Although Hawaiian hoary bat detectability at both Hakalau and Laupāhoehoe was notably lower during the outbreak year than in any year of the 5-year study conducted by Gorresen et al. (2013), this was likely due to the relatively short amount of foraging time required to reach a digestive bottleneck each night. Echolocation calls associated with searching and attacking insect prey peaked abnormally early in the night during the outbreak at Laupāhoehoe (no comparison of this type from Hakalau data is made because of insufficient bat calling activity). Bats actively foraged over longer portions of the night and presumably at lower success rates during non-outbreak times when moth densities were orders of magnitude lower. The timing of feeding among bat species varies relative to the activity and abundance of preferred prey species (Rydell et al., 1996). Furthermore, moth activity is strongly affected by temperature and wind with fewer moths collected in light traps on cold and windy nights (Jonason et al., 2014; Yela & Holyoak, 1997). We expect that koa moths were active well after sunset, based on personal observations and communications from others of moth activity at house lights many hours after sunset and “blizzards of moths crossing Saddle Road” (W.P. Haines, public communication in Stewart, 2013). With lower feeding efficiency, bats also are expected to have to call more (more call events and sound pulses) over longer portions of the night, which we observed during the non-outbreak years. Although the activity and frequency of feeding buzzes of some continental bat species increased when local densities of a pest moth were experimentally enhanced with synthetic sex pheromone lures, their behavior was not observed during a large-scale outbreak (Charbonnier et al., 2014).

During the 2013–2014 outbreak, Hawaiian hoary bats remained uncommon at Hakalau relative to their presence at Laupāhoehoe, as was observed by Gorresen et al. (2013) from 2007 to 2011. This is evidenced by the relative number of call events, echolocation pulses, and the detectability indices from the two locations. Thus, even during a period when koa moths were superabundant at Hakalau (see video link within https://doi.org/10.5066/P9HE9WKK), we did not detect an increase in bat foraging activity. Because koa moths were abundant and widespread during the 2013–2014 outbreak, there would have been little adaptive benefit to bats moving into Hakalau from surrounding areas. Some bat species track and exploit the local availability of prey within regional landscapes (McCracken et al., 2012), but aggregating behavior is more likely to occur in open habitats rather than in closed forest (Müller et al., 2012).

Response of parasitic wasps to outbreak

Hymenopteran parasitoids can drive moth population cycles in temperate, continental regions (Klemola et al., 2010; Turchin et al., 2003), but their role in tropical and insular environments is unclear. Koa moth outbreaks may at one time have occurred more frequently in the Hawaiian Islands (Haines et al., 2009; Perkins, 1913), but interactions between Hawaiian moths and their native predators and parasitoids are complicated, if not disrupted, by a diverse, abundant, and widespread invasive parasitoid community (Henneman & Memmott, 2001; Oboyski et al., 2004; Peck et al., 2008). Nevertheless, we expected to see increased numbers and attack rates of some members of this novel parasitoid community in response to the koa moth outbreak.

Consistent with previous reports (Davis, 1954; Haines et al., 2009; Zimmerman, 1958), H. exiguae, a long-established adventive parasitoid, was most frequently reared from koa moth caterpillars, and Meteorus laphygmae, introduced in 1942 to control sugarcane pests, also was reared. Both have broad host ranges and are widespread on Hawai‘i Island (Funasaki et al., 1988; Peck et al., 2008). Although malaise trap data indicate that both species tracked caterpillar abundance during the outbreak, the rate of parasitism (3.2%) was not proportional to caterpillar abundance. The rate was also less than that found during the late stage of the 2003–2004 Maui koa moth outbreak (20% [12 of 60], Haines et al., 2009), for caterpillars generally in Kaua‘i forests (20% [12 of 60], Henneman & Memmott, 2001), and for endemic Cydia spp. (Tortricidae) in subalpine woodland on Mauna Kea (39.5%; Oboyski et al., 2004). Our finding that 15% of pupae and pre-pupae at Laupāhoehoe were parasitized indicates that overall parasitism at Hakalau may be greater than the rate we observed only in caterpillars.

We did not assess mortality rates of koa moth eggs due to parasitism, predation, or pathogens, but predators can destroy many eggs of some outbreaking species (Castagneyrol et al., 2014). Host associations of egg parasitoids are poorly known in Hawai‘i, but the endemic Trichogramma semifumatum (Hymenoptera: Trichogrammatidae) has been reported to attack Scotorythra spp. eggs (Swezey, 1929). Other species of Trichogramma are often released as biocontrol agents of agricultural pests and can be a significant cause of Lepidoptera egg mortality (Smith, 1996). We could not attribute caterpillar mortality in the field or laboratory to pathogens, but entomopathogenic fungi or other factors may have gone undetected during the egg stage. Unidentified diseases were suspected of killing many caterpillars being reared in the laboratory during the 2003–2004 koa moth outbreak on Maui (Haines et al., 2009), and microorganisms may affect insect outbreak dynamics in a variety of complex ways and in concert with other factors (Cardoza et al., 2012; Elderd et al., 2013; Ewald, 1987; Páez et al., 2017; Royama et al., 2017).

It is unclear what level of parasitism might be necessary to suppress a koa moth outbreak, but non-native parasitoids did little to dampen the 2013–2014 irruption. Moreover, we found no evidence of parasitism by endemic species of Enicospilus (Ichneumonidae), despite earlier reports of Enicospilus attacking Scotorythra spp. (Swezey, 1929). Neither was Enicospilus reared from S. paludicola during the 2003–2004 outbreak on Maui, likely because Enicospilus attacks later instars than do H. exiguae and M. laphygmae (Haines et al., 2009). In an experimental study of the temperate, continental autumnal moth (Epirrita autumnata; Geometridae), a caterpillar parasitism rate of 88% was associated with a rapid decline in numbers to below initial outbreak levels (Klemola et al., 2010). Top-down factors may be important in limiting the frequency and scope of outbreaks (Dwyer et al., 2004; Letourneau, 2012; Ostfeld & Keesing, 2000; Symondson et al., 2002) even if they are not critical in stopping them, but quantifying the effects of native and non-native enemies on koa moth dynamics is challenging due to the unpredictable timing of outbreaks. Even in the intensively studied spruce budworm (Choristoneura fumiferana) outbreak system, Royama et al. (2017) were unable to identify a dominant effect of a single agent of mortality among the number they investigated.